基于小波包与Elman神经网络的整流电路故障诊断  

Research on Fault Diagnosis of Rectifier Circuits Based on Wavelet Packet and Elman Neural Network

作　　者：田质广[1] 赵刚[1] 

机构地区：[1]南京航空航天大学航空电源航空科技重点实验室,南京210016

出　　处：《系统仿真学报》2009年第10期2981-2984,共4页Journal of System Simulation

基　　金：中国博士后科学基金(2006040933)

摘　　要：引入了一种针对电励磁双凸极发电机的整流电路故障的小波和Elman神经网络相结合的诊断方法。根据整流电压波形的畸变规律,将2个以内的二极管故障分为7类共计22种故障元。选用Daubechies小波作为小波基,对实测的整流电压进行小波包分析,利用小波系数,在各个频带上进行小波信号重构。提取全频带能量特征值,并以此构造故障模式向量作为神经网络的输入,实现了故障的分类和定位。以一台840W、12/8极的电励磁双凸极发电机为实验样机,通过采集大量故障信号并进行信号处理,以故障特征值的容差范围作为误差判别标准,力求诊断误差最小化。实验结果表明,该方法对被分析信号的频率波动和幅值变化均具有良好的鲁棒性,是一种有效的故障诊断方法,具有较高的故障诊断正确率。A fault diagnosis algorithm for rectification circuit in Doubly salient electro-magnetic machine （DSEM） system based wavelet packet and Elman neural network was proposed. Then the faults （within 2 diodes faults） as 7 patterns or 22 fault states were classified according to distortion laws of rectified voltage waveforms. The wavelet coefficients were obtained from wavelet packet analysis of fault voltage signals, and then the wavelet reconstruction of each frequency band was processed. Poly-dimensional fault pattern vectors were established based on energy eigenvalue, energy eigenvalue is inputted Elman neural network, realized fault diagnosis and locating. In order to develop sample space of fault identification, taking 840W, 12/8 poles DSEG as a prototype, fault signals at great mass were gathered and then analysed through the methods mentioned above. Meanwhile, the energy-eigenvalue tolerance was introduced to minimize the diagnosis error. The results indicate that the fault analysis techniques are accurate and effective.

关 键 词：故障诊断 小波包分析 整流电路 能量特征值 ELMAN神经网络 

分 类 号：TM307.1[电气工程—电机]